Main content

Celonis drives process context to deliver meaningful enterprise AI

Derek du Preez Profile picture for user ddpreez June 9, 2024
Process mining vendor Celonis announced a swathe of AI updates to its platform, but most importantly, drives home the point about establishing context for AI.

Process Automation and Workflow Automation Concept - Automating the Flow of Tasks Across Work-related Activities in Accordance with Defined Business Rules © ArtemisDiana - Shutterstock
(© ArtemisDiana - Shutterstock)

Enterprises are having to think long and hard about their data to take advantage of the swathes of new AI technologies that are hitting the market. What’s becoming clear is that data locked in systems and silos, which don’t consider the full scope of enterprise operations, will quickly run into limitations. For a vendor like Celonis, which focuses on how processes run, regardless of the systems that underpin them, the opportunity lies in what it has called a ‘shared language’, to help AI models better deliver insights. 

And this week Celonis made a number of product announcements that further aim to make AI meaningful for enterprise buyers, where it sees process context as key. In recent months, Celonis has taken its process mining capabilities further, with the launch of its Process Intelligence Graph, which allows companies to develop a knowledge graph that maps an organizations’ ‘process knowledge’, and creates a ‘digital twin’ of how a business is running. This is made possible thanks to developments in Object Centric Process Mining, which allows organizations to map how different ‘objects’ (e.g. order, production, shipments, procurement) all interact with each other. 

It’s these interactions that provide useful context for AI systems. As Divya Krishnan, VP of Product Marketing at Celonis, explains: 

What's really missing is the context of how an organization really runs. There's no shortage of raw data sources for AI. The reason that ChatGPT can do things so effectively, like writing an invitation for your kids' birthday party, or suggesting an itinerary for a holiday in Provence, is because not only does it have this vast array of data sources ranging from books and articles and posts, but it also has sources like Wikipedia that are contextualizing it, that are structuring it, that are filling in the missing gaps. So it has the data, but it also has the context. 

Enterprises struggling to get highly meaningful knowledge out of their AI models may wish to think about the following: 

In the enterprise world, there's no limit to the raw data that's available, but there isn't anything that's giving that context. That's really what comes from the processes. I say context, I mean things like: how does your organization calculate on time delivery? What's the grace period? For some organizations, it can be three days, for others, it's five, for others it's two. There's nothing that's really capturing that for your enterprise and then giving it to AI. That's the kind of context that's totally missing. 

This is an important point. There will be many useful variations of AI in the enterprise - whilst some will carry out productivity tasks such as creating presentations or summarizing notes, getting to a nuts and bolts understanding of being able to ask AI how you can improve the running of your company is a much more sophisticated challenge. What that requires is some sort of meta or engagement layer that sits across enterprise processes, allowing users to ask about improvements, whilst understanding *why* things operate the way they do. 

As Krishnan notes, the connective understanding of why one process or outcome is tied to another, such as why scheduling errors in a system can lead to excess inventory and impact working capital, is a deeper problem to solve: 

I think ‘meta layer’ is a great way to put it. Very rightly, there's so much focus on getting your data ready for AI. But there's not much focus on what's going to translate and sit in between your data and AI. How do I make AI enterprise relevant for my organization? For it to be meaningful, and thus enterprise relevant, it has to have that understanding of how your organization operates and why. 

That business context, that process knowledge, that really stitches it together - above just the fact that all of these orders are happening or they're being canceled…What does it mean for something to be canceled? What does it mean for something to be confirmed and that's positive or negative? Is the payment term that's coming in favorable or unfavorable? 

I think we really take it for granted, but that's a lot of knowledge around how the organization runs - AI has to be fed with it, there's no other place that it's going to get that. If it doesn't have that, then it's always going to be capped in terms of how much impact it can have. It can only really go after a very limited array of use cases.

Product updates

Knowing that the outcome of Celonis’ product updates is this enterprise context to drive better AI outcomes, it made a number of announcements this week that include: 

  • Oracle EBS Transformations - users can now more easily bring Oracle data into the Process Intelligence Graph for key processes like Accounts Payable, Procurement, Order Management, Accounts Receivable, and Inventory Management.
  • Fast Data Model Search - organizations can more quickly find objects and events, whilst keeping searches focused with improved filters.
  • Improved Relationships UI - a new user interface has been rolled out to model objects and events, including the ability to show incoming or outgoing object relationships.
  • Intuitive Object Selection - using the PI Graph’s object-centric foundation, Celonis can now create cuts of the data for different use cases and users.
  • Celonis Process Management - a new suite of products including the Process Designer, Navigator, and Cockpit, which have been released following the integration of Symbio’s technology (acquired last year) into the Celonis platform.
  • Emporix Orchestration Engine - a new process orchestration engine that leverages Celonis' Process Intelligence to automate processes end-to-end in real-time.
  • Standard Data Ingestion API - again, focusing on data ingestion, organizations should now find it easier to get data into Celonis from source systems when using third-party data platforms and tools.
  • Premium Process Query Engine - for users, the Premium Process Query Engine can process up to 4X more data and will see processing speeds that are 3X faster.

In addition to the above, Celonis also highlighted a number of platform apps that target AI use cases for organizations. These include a Planning Parameter Optimization App, which providers material planners with updated planning parameters based on evolving consumption and replenishment patterns, as well as a Free-Text Requisition App, which improves spend under management by converting free-text requisitions to POs, as well as a Duplicate Invoice Checker, which addresses missed duplicate invoices help within ERP systems. 

Celonis also last year launched its Process Copilots, which allow customers to model their data into the Process Intelligence Graph and define the knowledge they are drawing on, which then allows them to select which KPIs, records and attributes the Copilots can access. Once this is complete, users across an organization can start asking questions, using this enterprise context. 

What’s significant to note here is that whilst Copilots offer a new way for users to engage with their organization, providing a certain level of abstraction away from the process (something that diginomica has spoken about previously), the process itself is still critical: 

A conversational agent, or a copilot, effectively operates as this abstraction layer away from the systems. But it's not really an abstraction away from the process, because the process is still running, there's just a different interface that somebody's actually engaging in, right? 

An order is still being routed, it's still being confirmed, it's still being blocked, it's still being unblocked, it's still being shipped. It's just a question of, where is somebody actually having to work for that to happen? And how easy is it? How much work do they have to do? How much effort do they have to put in? How much engagement did they have? And also how many systems did they have to hop across to actually be able to do it right? 

Generative AI in particular can play a massive role from a productivity perspective in making that much simpler, much easier. But the process is still central because it's the single thing that is in common across all systems and across all departments. It's about making sure that that can run as smoothly as efficiently in service of the outcomes that customers really want to achieve.

My take

Celonis is getting to the heart of an AI challenge here that I’ve been mulling a lot recently. Whilst a lot of the AI use cases we’ve seen from vendors in recent months focus on content (summarization, creation), these are largely low hanging fruit in the grand scheme of things. The much harder problem to solve is system-based AI, which pulls data from systems of record and understands how your organization is actually running. Some customers will do this via data clouds, such as Snowflake, and there will likely be siloed approaches within functional areas e.g. supply chain and HR. However, if you’re able to understand your processes organization-wide, then you’re likely to unlock much more meaningful outcomes. Being able to do this without relying on manual processes, or institutional knowledge locked in systems or held amongst your people, is a pretty compelling argument. And process sits at the center of all of that. 

A grey colored placeholder image